Getting started with WhereNext

Component 1: downloading and processing occurrences

The first step in running WhereNext is to define your study area. This may be a country, which you can select from the dropdown list, or a customized area, in which case you’d have to select the “User shape option” in module 1.1 to upload a zipped shapefile. Here, we’ll select Colombia from the list.

Next, you’ll need to select the source of your occurrence data in module 1.2. To download data from GBIF, select first the taxonomic rank in which you’re interested (class, order or family) and enter the name of your group. Here, we’ll choose “Family” and enter “Arecaceae” (without quotes, always capitalize the first letter) in the text box.

Then, click on the GBIF login button. You’ll need to enter your GBIF username, password and email. If you are not registered on GBIF go to https://www.gbif.org/ to do so. It is preferable that you sign up using the form instead of the quick registration methods (gmail, facebook, github), so that you know all of your credentials.

WhereNext will download the requested data from GBIF. This may take a few minutes. You can check on the download progress through the printed output of the R console or by logging into your GBIF profile and accessing your downloads. This is also a way to recover your occurrence data if for some reason the download is interrupted, or if you want to pre-process the data before uploading them manually through the “User-specified” option.

Once the download is finished, occurrences will be shown in the map and occurrence table tabs, and the citation for your download will be available in the results tab. By clicking on the colored clusters on the map, you may inspect individual occurrences. Further, you can access all record info available in GBIF by clicking on the ID link.

Finally, you may run the clean data function. This will run countries, capitals, centroids, equal, gbif, institutions, outliers, seas and zeros tests available in CoordinateCleaner. To learn more about these tests check the CoordinateCleaner package documentation (https://cran.r-project.org/web/packages/CoordinateCleaner/index.html).

Component 2: downloading and processing environmental layers

To download data from WorldCLim, all you need to do is to select the resolution of the analysis (2.5 min ~ 5 km, 5 min ~ 10 km, 10 min ~ 20 km), and click on the “Download WorldClim data” button. For dense sampling schemes, a higher resolution (e.g. 2.5 min or lower) may be desired. However, for ad hoc occurrence data, a lower resolution may be best as it will aggregate more occurrences, thus allowing a better estimation of turnover between cells. Hence, we will choose the 10 min resolution. Notice that you can also upload your own data by selecting the “User-specified” option in module 2.1. In real-world applications, you may want to upload your own set of variables that influence turnover in your occurrence data, such as vegetation and soil layers.

Once the download is completed, one variable will be plotted in the map tab. You may choose which variable is plotted by selecting it in the “Visualize layer” module. You can also reduce the redundancy in your environmental data by removing correlated variables. To do so, select a maximum correlation threshold and click on run in module 2.2. The selected variables will be shown on the list of layers to visualize and in the message box.

Component 3: estimating sampling completeness and running a GDM

We all the necessary inputs to run a GDM, which is the basis to compute the disimilarity from sampled to unsampled sites. However, with occurrece data from ad hoc surveys, there’s the real danger that some cells in the study area are not sufficiently sampled, thus inducing error in the estimation of disimilarity as a function of environment. The “Estimate cell stats” module, computes the number of species, sampling events (defined by default as days of sampling) and completeness based on the proportion of observed species richness and predicted species richness by the chao2 non-parametric estimator. Click on the run estimation to start this computation.

Once is finished, some red circles will be shown on top of the map. They represent cells that have been sampled on at least 2 occassions (the minimum to compute the chao2 estimator). Use the sliders in 3.2 to explore the consequence of using different criteria and thresholds to cell as “well sampled”. You can also look at the values of individual cells by hovering over the red dots on the map or downloading the cell stats file. There are no hard and fast rules to make this decision as it will depend on you study system, sampling methods used etc. For this exercise we will select sites that have been sampled at leat 5 times as our selection criterion. Once that selection has been made, click on the filter occurrences button, which will create a subset of occurrences to run the GDM.

Now, we’re ready to move to the “Configure GDM and run” module. To start this analysis, you will need to select a dissimilarity index from the list. Jaccard and Beta similarity are both recommended for presence-only data, but Beta similarity is richness independent (Lennon et al. 2001), which may be suitable for large areas. We will keep the options to include geographic distance as a predictor and perform further variable selection to remove uninformative variables. Click on the “Run GDM” button.

A GDM-derived spatial representation of disimilarity will be displayed in the map tab (uncheck the Env Data and Occ Data layers to see it properly). In this depiction of disimilarity, similar colors represent roughly similar biotic communities. For palms, we see a biotic differentiation between the lowlands and highlands, and within the lowlands there’s a distintion between the caribbean and llanos grasslands, while there’s little difference, at least visually, between the choco and amazonian tropical forests. You can further inspect this layer in GIS software by downloading it as a 3-band GeoTIFF file.

The results tab provides further info on the variables selected in the model and their coefficients as well as the deviance explained by the model. In this case, deviance explained was 39%, which could be further increasing by incorporating additional variables and/or improving criteria for cell selection.

Component 4: finding survey suggestions

By default, WhereNext suggest sites to sample anywhere in the study area. However, you may constraint the suggestions to a pre-defined list of sites by using the “From preselected sites option”. In this case, WhereNext will optimize the order in which sites are sampled. To use the “From preselected sites” option, you must upload a two column csv file with headers “decimalLongitude” and “decimalLatitude”.

For palms, we will find suggestions anywhere in the study area. To do so, first click on the “Compute complementarity button”. This will add a layer named “Complementarity” to the map (be sure to deactivate layers on top to see it) as well as the first suggestion by WhereNext represented by a light blue dot. After this, you can proceed iteratively, adding, rejecting, and modiying WhereNext suggestions, based on their suitability for sampling (e.g. some sites may have visiting restrictions, or be hard to reach, or not have currently suitable vegetation).

Finally, you may look at the reduction in sampling gaps by looking at the initED and outED columns on the results tab, which indicate how well the environmental space is being sampled. Lower values indicate better representation by existing survey locations and suggestions.